maddes8cht commited on
Commit
81d35d0
1 Parent(s): fe2124c

"Update README.md"

Browse files
Files changed (1) hide show
  1. README.md +7 -19
README.md CHANGED
@@ -9,21 +9,7 @@ I'm constantly enhancing these model descriptions to provide you with the most r
9
  - Model creator: [adept](https://huggingface.co/adept)
10
  - Original model: [persimmon-8b-base](https://huggingface.co/adept/persimmon-8b-base)
11
 
12
- # Important Update for Falcon Models in llama.cpp Versions After October 18, 2023
13
-
14
- As noted on the [Llama.cpp GitHub repository](https://github.com/ggerganov/llama.cpp#hot-topics), all new Llama.cpp releases after October 18, 2023, will require a re-quantization due to the new BPE tokenizer.
15
-
16
- **Good news!** I am glad that my re-quantization process for Falcon Models is nearly complete. Download the latest quantized models to ensure compatibility with recent llama.cpp software.
17
-
18
- **Key Points:**
19
-
20
- - **Stay Informed:** Keep an eye on software application release schedules using llama.cpp libraries.
21
- - **Monitor Upload Times:** Re-quantization is *almost* done. Watch for updates on my Hugging Face Model pages.
22
-
23
- **Important Compatibility Note:** Old software will work with old Falcon models, but expect updated software to exclusively support the new models.
24
-
25
- This change primarily affects **Falcon** and **Starcoder** models, with other models remaining unaffected.
26
-
27
 
28
 
29
 
@@ -35,19 +21,21 @@ The core project making use of the ggml library is the [llama.cpp](https://githu
35
 
36
  # Quantization variants
37
 
38
- There is a bunch of quantized files available. How to choose the best for you:
39
 
40
  # Legacy quants
41
 
42
  Q4_0, Q4_1, Q5_0, Q5_1 and Q8 are `legacy` quantization types.
43
  Nevertheless, they are fully supported, as there are several circumstances that cause certain model not to be compatible with the modern K-quants.
44
- Falcon 7B models cannot be quantized to K-quants.
 
 
45
 
46
  # K-quants
47
 
48
- K-quants are based on the idea that the quantization of certain parts affects the quality in different ways. If you quantize certain parts more and others less, you get a more powerful model with the same file size, or a smaller file size and lower memory load with comparable performance.
49
  So, if possible, use K-quants.
50
- With a Q6_K you should find it really hard to find a quality difference to the original model - ask your model two times the same question and you may encounter bigger quality differences.
51
 
52
 
53
 
 
9
  - Model creator: [adept](https://huggingface.co/adept)
10
  - Original model: [persimmon-8b-base](https://huggingface.co/adept/persimmon-8b-base)
11
 
12
+ Persimmon is a Large language Model from Adept AI. It is trained from Scratch with a context legth of 16k, which is 4 times the context size of LLaMA2 and 8 times that of GPT-3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
13
 
14
 
15
 
 
21
 
22
  # Quantization variants
23
 
24
+ There is a bunch of quantized files available to cater to your specific needs. Here's how to choose the best option for you:
25
 
26
  # Legacy quants
27
 
28
  Q4_0, Q4_1, Q5_0, Q5_1 and Q8 are `legacy` quantization types.
29
  Nevertheless, they are fully supported, as there are several circumstances that cause certain model not to be compatible with the modern K-quants.
30
+ ## Note:
31
+ Now there's a new option to use K-quants even for previously 'incompatible' models, although this involves some fallback solution that makes them not *real* K-quants. More details can be found in affected model descriptions.
32
+ (This mainly refers to Falcon 7b and Starcoder models)
33
 
34
  # K-quants
35
 
36
+ K-quants are designed with the idea that different levels of quantization in specific parts of the model can optimize performance, file size, and memory load.
37
  So, if possible, use K-quants.
38
+ With a Q6_K, you'll likely find it challenging to discern a quality difference from the original model - ask your model two times the same question and you may encounter bigger quality differences.
39
 
40
 
41